Autonomous operation method of power grid based on active adaptation of elastic charging load to active distribution

Abstract

The application discloses an active adaptation method of elastic charging load based on active distribution network autonomous operation, which comprises the following steps: collecting multi-source heterogeneous data, obtaining corresponding state estimation value, performing real-time voltage stability evaluation, and performing dynamic voltage stability index evaluation; through space-time prediction, virtual energy storage aggregation and elastic domain analysis, the adjustment capacity of the charging load is quantified; through multi-time scale collaborative optimization decision of day-ahead, day-in and real-time, the power adjustment instruction of each cluster corresponding to the virtual energy storage aggregation is generated; the instruction is decomposed and issued to each charging pile, so that the power adjustment instruction is adaptively executed by the local controller, and the execution deviation is monitored and corrected in real time; multi-dimensional quantitative evaluation and verification are performed, and the control strategy is continuously improved based on operation data. The system can accurately predict the voltage out-of-limit risk, and use the modeled elastic charging load resources for advance or real-time adjustment.

Description

Technical Field

[0001] This application relates to the field of power grid operation, specifically to a method for autonomous operation of a power grid based on the proactive adaptation of flexible charging loads to active distribution. Background Technology

[0002] Traditional voltage control methods suffer from passive lag and a "one-size-fits-all" approach. Their shortcomings lie in the fact that traditional methods often employ a "monitor-response" model, meaning that capacitor bank switching, transformer tap adjustment via on-load tap changers (OLTCs), and even the disconnection of distributed generation sources only occur after a voltage exceedance occurs. However, this response is delayed, and the adjustment methods are crude (e.g., resulting in large differences in capacitor bank switching levels), easily causing voltage oscillations or repeated adjustments. For voltage problems caused by rapid fluctuations in distributed generation (DG), the mechanical switching equipment cannot keep up with the speed.

[0003] To elaborate, the shortcomings of traditional solutions are mainly reflected in the following aspects: Regarding resource utilization: Electric vehicles are viewed as a "burden" rather than an "asset." The shortcomings are evident in the fact that traditional power distribution network planning and operation often treat electric vehicle charging loads simply as a new, rigid load that exacerbates peak-valley differences. To prevent the cumulative effect from causing equipment overload or voltage collapse, conservative management strategies such as restricting access and staggering peak power consumption are often adopted, or huge investments are required for grid upgrades. This not only inhibits the development of electric vehicles but also wastes the enormous load-side regulation potential.

[0004] Regarding the control architecture: there are issues of centralized rigidity and information silos. The shortcomings manifest in the fact that traditional Supervisory Control and Data Acquisition (SCADA) and Energy Management Systems (EMS) are highly centralized, requiring real-time data from the entire network to be aggregated at a main station for unified decision-making. This leads to high communication pressure, computational bottlenecks, and a high risk of single points of failure. Simultaneously, user-side data (such as charging plans and battery state of charge (SOC)) involves privacy concerns and is difficult to obtain completely transparently, creating "information silos" that hinder centralized optimization.

[0005] In terms of system development: It lacks learning and adaptive capabilities. The shortcomings manifest in the fact that traditional control systems often operate based on fixed models and preset rules. When changes occur in the power grid structure, distributed generation (DG) penetration rate, and load characteristics, their control parameters and strategies are often difficult to adjust automatically, requiring manual readjustment or upgrades, leading to high system maintenance costs and long-term performance degradation. Summary of the Invention

[0006] To address the aforementioned issues, this application proposes a method for autonomous grid operation based on flexible charging loads actively adapting to active distribution, comprising: Collect multi-source heterogeneous data, perform data fusion and power distribution network operation status estimation to obtain corresponding status estimates, and perform real-time voltage stability assessment and dynamic voltage stability index assessment based on the status estimates. Based on the user-side load data and external environment information contained in the multi-source heterogeneous data, the charging load regulation capability is quantified through spatiotemporal prediction, virtual energy storage aggregation and elastic domain analysis. Based on the state estimate, the voltage sensitivity matrix in the dynamic voltage stability index, the aggregated adjustable capacity of the virtual energy storage, and the migration range in time and space corresponding to the adjustment capability of the charging load, the power adjustment command of each cluster corresponding to the virtual energy storage aggregation is generated through day-day-intraday-real-time multi-timescale collaborative optimization decision-making. The power adjustment commands for each cluster are decomposed and sent to each charging pile so that the local controller can adaptively execute the power adjustment commands and monitor and correct the execution deviation in real time. Based on the measured voltage data of each charging pile during the actual execution of the power adjustment command, a multi-dimensional quantitative evaluation and verification are conducted, and the control strategy is continuously improved based on the operational data.

[0007] In one example, multi-source heterogeneous data is collected, and data fusion and distribution network operation status estimation are performed to obtain the corresponding status estimates, specifically including: The system collects grid operation parameters based on synchronous phasor measurement units, user-side load data based on smart meters, and distributed power source data based on distributed power source monitoring systems to obtain multi-source heterogeneous data. For the aforementioned multi-source heterogeneous data, a credibility-weighted fusion is performed to obtain fused data; Based on the weight diagonal matrix corresponding to the credibility weight, the measurement equation is established by weighted least squares method to obtain the corresponding optimal state variables; Based on the measurement function corresponding to the optimal state variable, the corresponding state estimate is obtained by recursive estimation through Kalman filtering dynamic estimation.

[0008] In one example, real-time voltage stability assessment and dynamic voltage stability index assessment are performed based on the state estimate, specifically including: Real-time voltage stability assessment is performed. Based on the values ​​in the state estimate, the corresponding Jacobian matrix is ​​determined. Based on the Jacobian matrix, the active power-voltage sensitivity matrix and the reactive power-voltage sensitivity matrix are obtained. Based on the active power-voltage sensitivity matrix and the reactive power-voltage sensitivity matrix, the voltage sensitivity matrix is ​​obtained. The dynamic voltage stability index is evaluated based on the voltage value in the state estimate. It allows voltage to perform over-limit tests within a set range and monitors the rate of change of voltage at key nodes and load response characteristics in real time during the over-limit process.

[0009] In one example, based on user-side load data and external environmental information contained in the multi-source heterogeneous data, the charging load regulation capability is quantified through spatiotemporal prediction, virtual energy storage aggregation, and elastic domain analysis, specifically including: Based on the user-side load data and external environment information contained in the multi-source heterogeneous data, user charging behavior is modeled through spatiotemporal prediction. The individual charging needs corresponding to the user charging behavior model are aggregated to obtain multiple clusters and corresponding virtual energy storage. Based on the user charging behavior modeling, elastic domain analysis is performed for each cluster to quantify the migration range of charging load in time and space. Based on the spatiotemporal distribution of the charging load corresponding to the user charging behavior model, the aggregate adjustable capacity of virtual energy storage, and the migration range of the charging load in time and space, the adjustment capability of the charging load is obtained.

[0010] In one example, based on user-side load data and external environmental information contained in the multi-source heterogeneous data, user charging behavior is modeled through spatiotemporal prediction, specifically including: Based on the user-side load data and external environment information contained in the multi-source heterogeneous data, a spatiotemporal prediction model is established through deep learning to predict the spatiotemporal distribution of charging load. Based on user location, battery state of charge, dwell time, and electricity price information from the external environment, a Markov decision process model is used to obtain the corresponding user sequence decision; and the user sequence decision is then corrected based on social attributes. Based on the changes in user charging power and the changes in electricity prices in the external environment, the corresponding demand elasticity coefficient is obtained.

[0011] In one example, the individual charging demands corresponding to the user charging behavior model are aggregated to obtain multiple clusters and corresponding virtual energy storage, specifically including: For each user's sequence decision-related individual charging demand, the charging load is divided into multiple clusters using a clustering algorithm; Based on the cluster partitioning results, the aggregate adjustable capacity is calculated; the aggregate adjustable capacity includes the upper and lower limits of aggregate power, the upper and lower limits of aggregate energy, and the energy time-series recursive relationship; Identify the virtual energy storage characteristic parameters of each cluster using historical data; For bidirectional charging and discharging scenarios, the discharge model is extended.

[0012] In one example, based on the user charging behavior modeling, elastic domain analysis is performed on each cluster to quantify the temporal and spatial migration range of the charging load, specifically including: Based on the individual charging needs, a feasible charging spatiotemporal domain is constructed; Based on the cluster partitioning results, a flexible resource map is drawn on the geographic information system platform to show the distribution of adjustable charging capacity in different regions and time periods. The potential for improvement of voltage distribution by spatiotemporal migration is quantitatively evaluated through heuristic algorithms. Construct a coupling matrix for inter-regional charging load migration to assess the feasibility and cost of transferring some charging demand to neighboring regions.

[0013] In one example, based on the state estimate, the voltage sensitivity matrix in the dynamic voltage stability index, the aggregated adjustable capacity of the virtual energy storage, and the temporal and spatial migration range corresponding to the charging load's adjustment capability, a multi-timescale collaborative optimization decision-making process (day-to-day-real-time) is used to generate power adjustment commands for each cluster corresponding to the virtual energy storage aggregation. Specifically, this includes: Based on the aggregate adjustable capacity of the virtual energy storage and the migration range in time and space corresponding to the adjustment capability of the charging load as constraint boundaries, a day-ahead stochastic optimization model is constructed, and a baseline scheduling plan is formulated; an intraday rolling optimization model is formulated through a model predictive control framework; a real-time feedback correction model is constructed to adjust power based on real-time voltage deviation; and a reverse scheduling scenario is extended to proactively increase the charging load to absorb renewable energy when the output of distributed power sources is excessive. Based on the voltage sensitivity matrix in the dynamic voltage stability index, the corresponding upper and lower voltage limits are transformed into feasible domain boundaries for charging power adjustment, and a time-varying voltage safety boundary is constructed; and a constraint relaxation and penalty mechanism is introduced. Based on the cluster partitioning results of the virtual energy storage, the global optimization problem is decomposed into subproblems of each cluster using the alternating direction multiplier method for parallel solution; and the consensus optimization algorithm is used to exchange boundary information among the cluster controllers to iteratively achieve the global optimum; and a federated learning mechanism is introduced to protect user privacy by ensuring that each cluster does not share original user data and only uploads model gradients or optimization boundary parameters.

[0014] In one example, the power adjustment command for each cluster is decomposed and sent to each charging pile, so that the local controller can adaptively execute the power adjustment command and monitor and correct the execution deviation in real time. Specifically, this includes: For the power adjustment commands of each cluster, the total adjustment amount of the aggregation layer is decomposed into the single power command of each charging pile through a power allocation algorithm based on fairness, and sent to each charging pile through an encryption algorithm; the elastic weight is dynamically adjusted according to the user's active settings; and a multi-level priority strategy is designed to handle the conflict between emergency control and regular scheduling. The local controller, based on the issued power command of the single pile and the locally collected user target state of charge, performs local optimization through model predictive control to obtain the corresponding power data; and automatically switches to islanded autonomous mode when communication is interrupted, and performs power self-balancing based on local voltage-frequency droop characteristics and energy storage status. The local controller executes locally optimized power data, detects the deviation between actual power and commanded power, and triggers adaptive correction when the deviation exceeds the allowable error threshold. It also handles communication interruptions and equipment failures according to the preset abnormal situation handling plan.

[0015] In one example, based on the measured voltage data of each charging pile during the actual execution of the power adjustment command, a multi-dimensional quantitative evaluation and verification are performed, and the control strategy is continuously improved based on the operational data, specifically including: Based on the measured voltage data of each charging pile during the actual execution of the power adjustment command, the voltage deviation index, voltage qualification rate, and the reduction ratio of voltage fluctuation rate before and after control are statistically analyzed to identify and improve the corresponding nodes; and based on the response records of the over-limit test, a dedicated evaluation index is constructed. Based on the power grid operation data and user charging data before and after the execution of the power adjustment command, the corresponding revenue data is calculated through comprehensive cost-benefit analysis; user subjective feelings are quantified into a satisfaction index through a questionnaire; and the peak-valley difference rate and load factor are calculated to evaluate the optimization degree of the load curve. Based on historical operating data and a library of typical scenarios, multiple test scenarios are randomly generated through Monte Carlo simulation to test the success rate of the power regulation command; extreme scenarios are simulated to verify the system reliability; and a social disturbance scenario is introduced to test the adaptability of the power regulation command. Based on historical operational data, association rule mining, cluster analysis, causal inference, and temporal clustering Hidden Markov Model are used to extract typical scenario libraries, control laws, and user behavior evolution paths. The control strategy is iteratively updated through reinforcement learning. The model parameters are updated in real time and the elastic contract incentive standard is dynamically adjusted through online learning algorithms.

[0016] The grid autonomous operation method based on the active adaptation of flexible charging load to active distribution proposed in this application can bring the following benefits: 1. Improve the voltage safety and power quality of the distribution network: Through real-time sensing and evaluation and multi-scale optimization decision-making, the system can accurately predict the risk of voltage exceeding the limit and make advance or real-time adjustments using modeled elastic charging load resources.

[0017] 2. Enhance the system's ability to absorb high proportions of renewable energy: Traditional distribution networks often have limited regulation capacity and need to adopt measures such as curtailing solar and wind power to ensure safety, resulting in the waste of clean energy. This application transforms the massive electric vehicle charging load into "virtual energy storage," providing bidirectional regulation space.

[0018] 3. Achieving economic operation of the power grid and win-win situation for all parties: By optimizing the model, pursuing economic optimization under the premise of ensuring safety, a new value stream is created.

[0019] 4. Build a smart distribution network ecosystem with self-learning and self-adaptive capabilities: knowledge accumulation and adaptive optimization enable the system to go beyond the scope of static optimization and have the ability to continuously evolve. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart illustrating the autonomous operation method of the power grid based on the active adaptation of flexible charging load to active distribution in the embodiments of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0023] like Figure 1 As shown, this application provides a method for autonomous grid operation based on flexible charging loads actively adapting to active distribution, including: S101: Collect multi-source heterogeneous data, perform data fusion and power distribution network operation status estimation to obtain corresponding status estimates, and perform real-time voltage stability assessment based on the status estimates to obtain dynamic voltage stability indicators.

[0024] Step S101 is mainly used to establish a real-time digital mirror of the distribution network's operating status, providing an accurate and comprehensive data foundation for subsequent decision-making.

[0025] Specifically, grid operating parameters are collected based on synchronous phasor measurement units, user-side load data are collected based on smart meters, and distributed power source data are collected based on distributed power source monitoring systems to obtain multi-source heterogeneous data.

[0026] High-precision, time-synchronized power grid operating parameters are obtained by acquiring data from multiple heterogeneous sources, ensuring the accuracy and timeliness of the sensing data. These power grid operating parameters include, but are not limited to, voltage phase angle, amplitude, and frequency.

[0027] For the acquisition of power grid operating parameters: a synchronous phasor measurement unit (PMU) is deployed, and PMU devices are installed at key nodes to acquire data such as voltage phase angle, amplitude, and frequency at a frequency of 50-60 frames per second, with time synchronization accuracy reaching the microsecond level.

[0028] Synchronous phasor measurement units are key equipment in the field of dynamic monitoring and control of power systems. Based on high-precision synchronous clock signals provided by the Global Positioning System (GPS) or the BeiDou Navigation Satellite System, they synchronously sample electrical quantities such as voltage and current in the power system.

[0029] The PMU (Power Controller Unit) can measure power grid operating parameters such as amplitude, phase, and frequency of electrical quantities in real time with extremely high accuracy and synchronization, and represent and transmit this data in the form of synchronization phasors. Unlike traditional measuring devices, its synchronization accuracy can reach the microsecond level, enabling synchronous measurement of electrical quantities at different locations over a wide area.

[0030] The Wide Area Measurement System (WAMS) built using the PMU can acquire real-time dynamic information about the entire power system, providing dispatchers with comprehensive and accurate system operating status. This helps to promptly identify potential problems in the system, such as low-frequency oscillations and voltage instability, and then take effective measures to ensure the safe and stable operation of the power system, thereby improving the intelligence level of the power grid and its ability to respond to emergencies.

[0031] The IEEE C37.118.2 communication protocol can be used to achieve standardized PMU data transmission, and fiber optic Ethernet or a 5G private network can be used to ensure the real-time performance and reliability of data transmission. Data timestamps are unified to Coordinated Universal Time (UTC) to eliminate clock skew between different devices.

[0032] For user-side load data collection: Advanced Metering Infrastructure (AMI) is used to collect user-side load data, such as real-time power, current, and voltage information of charging piles.

[0033] For the acquisition of distributed power source data: Real-time data such as active / reactive power output and operating status of distributed power sources such as photovoltaics and wind turbines are obtained through the communication interface between the SCADA system and the inverter.

[0034] For multi-source heterogeneous data, a credibility-weighted fusion is performed to obtain fused data. Credibility-weighted fusion is applied to data from different sources and with different levels of precision. The credibility of each source's data can be determined based on the precision level and reliability indicators of the measurement equipment. For example, corresponding reliability indicators are pre-set for different measurement equipment (including PMU, SCADA systems, and equipment corresponding to advanced measurement systems). Simultaneously, based on the precision level of each measurement equipment, the data is normalized and weighted by the reliability indicators, and then summed to obtain the corresponding credibility weight.

[0035] Based on the weight diagonal matrix corresponding to the confidence weights, the measurement equation is established using the weighted least squares method to obtain the corresponding optimal state variables. This is mainly used to eliminate measurement errors, reconstruct the complete operating state of the distribution network, and identify adverse data.

[0036] When performing weighted least squares state estimation, the optimal state variables are solved by establishing measurement equations, as shown in Formula 1: Formula 1; in, It is a state variable vector containing the voltage magnitude and phase angle of all nodes, which can be obtained based on the power grid operating parameters; The measurement vector contains the measured values ​​of power, voltage, current, etc. at all measurement points, which can be obtained based on the actual measured values ​​of each data point in the fused data. For measurement functions, describe the physical relationship between state variables and measured values; This is a weighted diagonal matrix that reflects the accuracy and reliability of different measurement devices.

[0037] Formula 1 obtains the state estimate that is most likely to be close to the true state by minimizing the weighted sum of squares of the measurement residuals, effectively handling measurement noise and redundant information.

[0038] Based on the measurement function corresponding to the optimal state variable, the corresponding state estimate is obtained through recursive estimation using Kalman filter dynamic estimation. Kalman filter dynamic estimation can perform recursive estimation for time-varying systems, adapting to the dynamic changes of the distribution network, as shown in Formula 2: Formula 2; in, This is the state estimate at time k; The error covariance matrix is ​​used to estimate the accuracy of the estimation. The Kalman gain matrix determines the degree to which the new measurement corrects the estimate; This is the process noise covariance matrix, reflecting the uncertainty of the system model; The noise covariance matrix is ​​used to measure measurement uncertainty; This is the state transition matrix, which describes how the system state evolves over time. To control the input matrix; The measurement matrix (the matrix form of the measurement function) describes the relationship between the state and the measured values. It is an identity matrix.

[0039] Formula 2 achieves optimal estimation in dynamic systems and is particularly suitable for scenarios where the output of distributed power sources fluctuates rapidly.

[0040] Furthermore, in addition to conventional state estimation, Generative Adversarial Networks (GANs) are introduced to simulate and identify potential data attacks, enhancing the system's robustness against malicious data injection. By constructing an "attack-defense" adversarial training mechanism, the state estimation model can maintain reliable estimation accuracy even when some measurement data is tampered with or communication is interrupted.

[0041] Real-time voltage stability assessment is performed by determining the corresponding Jacobian matrix based on the values ​​in the state estimate. Based on the Jacobian matrix, the active power-voltage sensitivity matrix and the reactive power-voltage sensitivity matrix are obtained, and based on these matrices, the voltage sensitivity matrix is ​​derived. Real-time voltage stability assessment is primarily used to quantify the current voltage stability margin of the power grid and identify weak points and risk areas.

[0042] The voltage sensitivity matrix is ​​calculated as shown in Formula 3: Formula 3; Where S is the sensitivity matrix, as shown in Formula 4. Formula 4; in, This is the node voltage change vector; This is the vector of active power change at the node; The vector of reactive power change at the node; This is the active power-voltage sensitivity matrix, which describes the effect of active power changes on voltage. This is the reactive power-voltage sensitivity matrix, which describes the effect of reactive power changes on voltage. Let be the Jacobian matrix, the linearization coefficient matrix of the power flow equations; This is the node voltage phase angle vector.

[0043] Formula 3 quickly assesses the impact of load or power changes on node voltage, providing a quantitative basis for control decisions. Dynamic voltage stability indices can be obtained based on voltage values ​​from state estimates.

[0044] Dynamic voltage stability index can be calculated using... - The voltage collapse proximity indicator (VCPI) can be used for online evaluation, as shown in Formula 5: Formula 5; in, Let be the voltage stability index of node j, with a value range of [0,1], where 0 indicates absolute stability and 1 indicates voltage collapse; The element in the j-th row and i-th column of the nodal admittance matrix; Let be the voltages at nodes i and j, respectively, which can be obtained through state estimation. This represents the total number of system nodes.

[0045] Formula 5 provides an intuitive quantitative indicator of voltage stability, making it easy to set control thresholds.

[0046] Of course, dynamic voltage stability indicators can also be predicted using a rapid evaluation model based on machine learning. By inputting real-time measurement data, dynamic voltage stability indicators can be directly output.

[0047] The system allows voltage to exceed limits within a set range, and monitors the rate of change of voltage at key nodes and load response characteristics in real time during this process. Additionally, provided system operating conditions permit and a safety pre-assessment has been conducted, the voltage is briefly allowed to exceed limits within the set range (e.g., not exceeding 5% of rated voltage, duration not exceeding 30 seconds) to proactively stimulate system response and test the rapid response capability of regulatory resources. This mechanism can be used to identify control dead zones, verify the dynamic adaptability of control strategies, and serve as a proactive means of system robustness testing. Real-time monitoring of the rate of change of voltage at key nodes and load response characteristics during limit exceedances forms a closed-loop testing process of "proactive disturbance - response assessment - parameter correction."

[0048] S102: Based on the user-side load data and external environment information contained in the multi-source heterogeneous data, the adjustment capability of the charging load is quantified through spatiotemporal prediction, virtual energy storage aggregation and elastic domain analysis.

[0049] This step S102 quantifies the adjustability by accurately characterizing the spatiotemporal distribution characteristics of the charging load.

[0050] Specifically, the first step involves modeling user charging behavior based on user-side load data contained in multi-source heterogeneous data, as well as external environmental information, through spatiotemporal prediction. This is primarily used to predict the spatiotemporal distribution of charging load in future periods and to identify rigid and elastic demand.

[0051] Based on user-side load data contained in multi-source heterogeneous data, as well as external environmental information, a spatiotemporal prediction model is established through deep learning to predict the spatiotemporal distribution of charging load. The model architecture can adopt a Convolutional Long Short-Term Memory Network (ConvLSTM) or a spatiotemporal graph neural network. Input features include multi-dimensional data such as historical charging data, user attributes, weather, electricity price, and traffic flow. Among them, user attributes, weather, electricity price, and traffic flow belong to external environmental information. Output data: the probability distribution of charging power in each area within the next 1-24 hours.

[0052] Based on user location, battery state of charge, dwell time, and electricity price information from the external environment, a Markov decision process model is used to obtain the corresponding user sequence decision.

[0053] The Markov decision process model is shown in Equation 6: Formula Six; in, This is the state space, which includes user location, battery SOC, dwell time, etc. This provides the operational space, including charging power levels and start / stop decisions; Let be the state transition probability function. This represents the probability of transitioning to state s' after performing action a in state s; The reward function considers multiple objectives, including user satisfaction and power grid revenue. As a discount factor, it balances the importance of immediate rewards versus future rewards.

[0054] Formula 6 models user charging behavior as a sequential decision-making process, which facilitates the analysis of user responses to control strategies.

[0055] This model refines user sequence decisions based on social attributes. It incorporates social network influence factors into charging behavior prediction, constructing a social-charging behavior coupling model based on graph neural networks. The model considers the influence of users' travel sharing, charging experience evaluations, and group activities (such as gatherings and events) on the spatiotemporal distribution of charging demand on social platforms. A "social incentive" mechanism is introduced, allowing users to earn social points, leaderboard rankings, or community identity badges by participating in charging regulation, thus enhancing user participation. This model expands charging behavior from individual rational decision-making to group social behavior, improving prediction accuracy and regulatory acceptance.

[0056] Based on the changes in user charging power and changes in electricity prices in the external environment, the corresponding demand elasticity coefficient is obtained. It can be represented as shown in Formula 7: Formula 7; in, Price elasticity of user i's charging demand during time period t; The amount of electricity charged for user i during time period t; The change in the amount of electricity charged by user i during time period t; Let t be the electricity price for period t; Let t be the change in electricity price during the time period t.

[0057] Formula 7 quantifies the degree of user response to price signals, providing a basis for elastic scheduling.

[0058] The second step involves aggregating the individual charging demands corresponding to the user charging behavior models to obtain multiple clusters and corresponding virtual energy storage. This is primarily used to aggregate dispersed charging loads into uniformly dispatchable virtual energy storage resources.

[0059] For each user's sequence decision corresponding to individual charging demand, the charging load is divided into multiple clusters using a clustering algorithm. Based on the distributed aggregation of cluster partitioning, K-means or spectral clustering algorithms can be used to divide the charging load into multiple clusters according to geographical location and adjustability characteristics, with each cluster managed by a local controller.

[0060] Based on the cluster partitioning results, the aggregate adjustable capacity is calculated. The aggregate adjustable capacity includes the upper and lower limits of aggregate power, the upper and lower limits of aggregate energy, and the energy time-series recursive relationship. The aggregate adjustable capacity calculation is shown in Formula 8. Formula 8; in, The total power of the aggregated charging load at time t; These are the minimum and maximum values ​​of the polymerization power at time t, respectively; The total energy of the aggregated charging load at time t; The charging efficiency coefficient takes into account energy loss during the charging process. For time intervals; These are the minimum and maximum values ​​of the polymerization energy, respectively.

[0061] Formula 8 aggregates the charging flexibility of a large number of electric vehicles into a continuously adjustable resource similar to energy storage.

[0062] The virtual energy storage characteristic parameters of each cluster are identified using historical data. The equivalent charge / discharge efficiency, self-discharge rate, and cycle life of the aggregate are also identified using historical data.

[0063] For bidirectional vehicle-to-grid (V2G) scenarios, the discharge model is extended. In V2G-supported scenarios, the virtual energy storage model is extended to include discharge capacity, defining the discharge depth, cycle life loss, and user revenue models for the aggregate. A "discharge willingness coefficient" is introduced, dynamically adjusting the adjustable discharge capacity based on the user's real-time electricity price response, battery state of health (SOH), and historical discharge behavior. This model supports the aggregate actively discharging during periods of high grid load, participating in frequency regulation or voltage support, thus upgrading from "elastic load" to "bidirectional elastic resource."

[0064] The third step involves modeling user charging behavior and conducting elastic domain analysis for each cluster to quantify the migration range of charging load in time and space.

[0065] Based on individual charging needs, a feasible spatiotemporal domain for charging is constructed. The spatiotemporal elastic domain is constructed as shown in Equation Nine: Formula 10; in, The feasible charging time and space domain for user i, including all feasible charging time and power combinations; The time it takes for the user to arrive at the charging station; The time the user leaves the charging station; To meet user demand for charging capacity; This refers to the charging power. These are the minimum and maximum power of the charging station, respectively; This represents the initial state of charge of the battery. The target state of charge; The minimum permissible state of charge; Battery capacity; For charging efficiency.

[0066] Formula 10 precisely describes the charging flexibility boundary for each electric vehicle.

[0067] Based on the cluster partitioning results, a flexible resource map is drawn on a Geographic Information System (GIS) platform to display the distribution of adjustable charging capacity in different regions and time periods. A heuristic algorithm is used to quantitatively evaluate the potential of spatiotemporal load migration to improve voltage distribution. The distribution of adjustable charging capacity in different regions and time periods is visualized on the GIS platform. A heuristic algorithm is developed to evaluate the potential of load spatiotemporal migration to improve voltage distribution.

[0068] A coupling matrix for inter-regional charging load migration is constructed to assess the feasibility and cost of transferring some charging demand to neighboring areas. Based on the conventional spatiotemporal elastic domain, a cross-regional collaborative migration mechanism is introduced, allowing users to transfer or exchange "charging rights" among multiple charging stations. By constructing the inter-regional charging load migration coupling matrix, the feasibility and cost of transferring some charging demand to neighboring areas when voltage exceeds limits or transformers are overloaded in a certain area are evaluated. This mechanism is particularly suitable for urban clusters or highway service area scenarios, improving the overall system resilience.

[0069] Based on steps one through three of S102 above, the spatiotemporal distribution of charging load, the aggregated adjustable capacity of virtual energy storage, and the migration range of charging load in time and space corresponding to the user charging behavior model are used to obtain the adjustment capability of charging load.

[0070] S103: Based on the state estimate, the voltage sensitivity matrix in the dynamic voltage stability index, the aggregated adjustable capacity of the virtual energy storage, and the migration range in time and space corresponding to the adjustment capability of the charging load, the power adjustment command of each cluster corresponding to the virtual energy storage aggregation is generated through day-day-intraday-real-time multi-timescale collaborative optimization decision-making.

[0071] This step S103 formulates the optimal charging load scheduling strategy to balance grid security, economic operation, and user demand.

[0072] Specifically, the first step is to construct a two-stage optimization model for the day-to-day and intraday periods, and establish a multi-timescale optimization framework that takes into account uncertainties.

[0073] Based on the aggregate adjustable capacity of virtual energy storage and the time and space migration range corresponding to the adjustment capability of charging load as constraint boundaries, a day-ahead stochastic optimization model is constructed and a baseline scheduling plan is formulated; and an intraday rolling optimization model is formulated through a model predictive control framework.

[0074] The current stochastic optimization model is shown in Equation 11: Formula 11; in, For the expectation operator, consider distributed generation and load forecasting errors; This represents the expected total cost. The unit cost of purchasing electricity from the main grid during time period t; The power purchased from the main grid during time period t; The unit cost of network loss during time period t; The power loss during time period t; Let t be the compensation cost for user i during time period t; To optimize the total number of time periods; Output of distributed power sources during time period t; This represents the normal load power during time period t. The charging load power during time period t; Let be the voltage of node i during time period t; These are the lower and upper voltage limits, respectively. Let be the current in line l during time period t; This represents the upper limit of the current in line l.

[0075] Formula 11 formulates the economically optimal scheduling plan in the day-ahead phase, taking into account forecast uncertainty.

[0076] The intraday rolling optimization model employs a Model Predictive Control (MPC) framework, which, for example, re-optimizes the scheduling plan for the next 1-4 hours every 15 minutes using the latest forecast and measured data.

[0077] Meanwhile, based on the two-stage optimization model of day-day and intraday, a real-time feedback correction model is constructed to form a collaborative optimization decision-making system with multiple time scales of day-day, intraday, and real-time, and power adjustment is performed based on real-time voltage deviation; the real-time feedback correction model adjusts power based on real-time voltage deviation at the second-level time scale.

[0078] The reverse scheduling scenario is expanded to proactively increase charging load to absorb renewable energy when distributed power generation output is excessive. The optimization model for the reverse scheduling scenario is extended: based on the conventional scheduling objective, a "reverse scheduling" scenario is added: when distributed power generation output is excessive and the main grid's peak-shaving capacity is insufficient, the charging load is proactively increased (even forced charging) to absorb renewable energy. In this scenario, the optimization objective shifts from "peak shaving and valley filling" to "proactive valley filling," and a renewable energy curtailment penalty term and a user forced charging compensation term are introduced into the objective function. The model must simultaneously consider battery overcharging risks and user scheduled departure time constraints to ensure that reverse scheduling does not impair users' basic travel needs.

[0079] The second step is to transform the voltage safety constraint into a feasible domain constraint for power regulation based on the target-constraint transformation of voltage sensitivity.

[0080] Based on the voltage sensitivity matrix in the dynamic voltage stability index, the corresponding upper and lower voltage limits are transformed into feasible domain boundaries for charging power regulation, and a time-varying voltage safety boundary is constructed. The feasible power regulation domain is calculated as shown in Equations XII and XIII: Formula 12; Formula Thirteen; in, The charging power adjustment amount for time period t; This represents the minimum allowable power adjustment during time period t. This represents the maximum allowable power adjustment during time period t. Let be the actual voltage value of node i during time period t; These are the lower and upper voltage limits, respectively. Voltage sensitivity of the charging load at node i; The planned power of the charging load during time period t; These represent the minimum and maximum allowable power of the charging load during time period t, respectively.

[0081] Formulas 12 and 13 transform the upper and lower voltage limits into upper and lower limits of power regulation, simplifying the optimization problem.

[0082] Simultaneously, a voltage safety funnel design is implemented to construct a time-varying voltage safety boundary, ensuring that the voltage does not exceed the limit under any circumstances.

[0083] By introducing constraint relaxation and penalty mechanisms, relaxation variables and penalty terms are introduced for non-critical constraints to ensure the feasibility of the optimization problem.

[0084] The third step is to design a distributed collaborative solution algorithm, and develop an efficient and scalable solution algorithm to adapt to large-scale problems.

[0085] Based on the cluster partitioning results of virtual energy storage, the global optimization problem is decomposed into subproblems for each cluster using the Alternating Direction Method of Multipliers (ADMM) for parallel solution. The Alternating Direction Method of Multipliers (ADMM) is shown in Equation XIV: Formula Fourteen; in, For dual variables (Lagrange multipliers); The penalty parameter controls the update step size of the dual variable; To augment the Lagrange function; This represents the number of iterations. These are the coefficient matrix and constant term of the constraint equation, respectively.

[0086] Formula 14 decomposes large-scale optimization problems into multiple subproblems that are solved in parallel, making it suitable for distributed computing architectures.

[0087] A consensus optimization algorithm is used to exchange boundary information among the cluster controllers to iteratively reach the global optimum. Each charging cluster controller only exchanges a small amount of boundary information, and iteratively reaches the global optimum.

[0088] A federated learning mechanism is introduced to protect user privacy by ensuring that each cluster does not share raw user data, but only uploads model gradients or optimization boundary parameters. Intelligent decision-making based on deep reinforcement learning can train agents to learn optimal scheduling strategies and adapt to complex dynamic environments. In a distributed solution architecture, the federated learning mechanism is introduced so that each charging cluster controller does not share raw user data, but only uploads model gradients or optimization boundary parameters. The central server aggregates and distributes the global model. This mechanism achieves global collaborative optimization while protecting user privacy, and is particularly suitable for cross-operator and cross-regional charging resource collaborative scheduling scenarios.

[0089] S104: The power adjustment command for each cluster is decomposed and sent to each charging pile so that the local controller can adaptively execute the power adjustment command and monitor the execution deviation in real time for correction.

[0090] This step is mainly used to translate optimization decisions into actual control actions, ensuring the accurate and reliable execution of instructions.

[0091] Specifically, for the power adjustment commands of each cluster, a fairness-based power allocation algorithm is used to decompose the total adjustment amount of the aggregation layer into individual power commands for each charging pile, which are then sent to each charging pile using an encryption algorithm. The control command decomposition and distribution process further decomposes the optimization commands of the aggregation layer into executable commands for individual charging piles.

[0092] The power allocation algorithm based on fairness is shown in Formula 15: Formula 15; in, The actual charging power of user i during time period t; The basic power requirement for user i; This represents the total power that the polymerization layer needs to adjust during time period t. For user i's elastic power; The base power allocation factor for user i; The elastic power allocation coefficient for user i; This represents the total elastic power consumption for all users.

[0093] Formula 15 ensures fairness among users while meeting the needs of the power grid.

[0094] Meanwhile, for command encryption and secure transmission, the national cryptographic SM2 / SM4 algorithm can be used to encrypt control commands and transmit them via HTTPS / MQTT protocol.

[0095] The elastic weights are dynamically adjusted based on user-defined preferences. Instruction priority and conflict resolution mechanisms: A multi-level priority strategy is designed to handle conflicts between emergency control and routine scheduling.

[0096] A multi-level priority strategy is designed to handle the conflict between emergency control and routine scheduling. A user-initiated intervention and preference priority mechanism is implemented. A user-initiated intervention interface is added to the power allocation model, allowing users to set preferences such as "whether to accept scheduling," "whether to participate in V2G," and "minimum off-site SOC" via mobile devices. The system dynamically adjusts its elastic weights accordingly. A "preference priority queue" is introduced, where only basic protection control is performed on users who explicitly refuse scheduling, while users with high preferences are given higher elastic weights, improving user acceptance and system control flexibility.

[0097] The second step involves using the local controller to perform local optimization based on the issued power command for each charging pile and the locally collected target state of charge of the user, through model predictive control, to obtain the corresponding power data. The local controller's adaptive response allows it to execute control based on received commands and local conditions.

[0098] The model predictive control local optimization is shown in Equation Sixteen: Formula Sixteen; in, Optimize the objective function locally; This refers to the actual charging power at the local location. This is a reference power command issued by the superior authority. This indicates the battery's current state of charge. The target state of charge; These are weight matrices, balancing tracking accuracy with user needs; It is the square of the weighted L2 norm.

[0099] Formula 16 optimizes local control performance while satisfying instructions from higher authorities.

[0100] At the same time, a plug-and-play protocol is set up, adopting the ISO 15118 / IEC 63110 standard, to realize automatic identification and protocol matching between electric vehicles and charging piles.

[0101] Simultaneously, local safety protection logic is set: protection thresholds for overcurrent, overvoltage, and leakage are set to ensure charging safety.

[0102] In the event of a communication interruption, the system automatically switches to islanded autonomous mode, performing power self-balancing based on local voltage-frequency droop characteristics and energy storage status. In the event of a communication interruption or master station disconnection, the local controller automatically switches to islanded autonomous mode, performing power self-balancing based on local voltage-frequency droop characteristics and energy storage status. The controller incorporates lightweight state estimation and local optimization algorithms to ensure that node voltage and frequency remain within safe boundaries even without upper-level commands, and records data from abnormal periods for post-event analysis and strategy correction.

[0103] The third step involves executing the locally optimized power data via the local controller, detecting the deviation between the actual power and the commanded power, and triggering adaptive correction when the deviation exceeds the allowable error threshold. Real-time monitoring and correction throughout the execution process are used to monitor the effectiveness of control commands and promptly correct deviations.

[0104] The deviation detection algorithm is shown in Formula 17: Formula 17; in, Let t be the power tracking error at time t; The actual power at time t; The command power at time t; This is the allowable error threshold.

[0105] It also sets up adaptive proportional-integral-derivative (PID) correction control, which dynamically adjusts the PID parameters according to the deviation to improve tracking accuracy.

[0106] Handle communication interruptions and equipment malfunctions according to the pre-set contingency plan. The contingency plan outlines the procedures for handling abnormal situations such as communication interruptions and equipment malfunctions.

[0107] S105: Based on the measured voltage data of each charging pile during the actual execution of the power adjustment command, a multi-dimensional quantitative evaluation and verification are performed.

[0108] This step is mainly used to quantify the effect of the control strategy and verify the effectiveness of the method.

[0109] Specifically, based on the measured voltage data of each charging pile during the actual execution of power adjustment commands, the voltage deviation index, voltage qualification rate, and the reduction ratio of voltage fluctuation rate before and after control are statistically analyzed to identify and improve the corresponding nodes.

[0110] The voltage deviation statistical index is calculated as shown in Formula 18: Formula 18; in, The voltage non-compliance index at time t; Let be the voltage at node i at time t; Rated voltage; This represents the total number of monitoring nodes. Voltage qualification rate; This represents the total duration of voltage over-limit operation. This represents the total runtime.

[0111] Simultaneously, the voltage fluctuation mitigation effect is quantified, calculating the reduction percentage of voltage fluctuation rate before and after control. Furthermore, a weak node improvement analysis is conducted to identify the nodes with the most and least significant improvements and to analyze the reasons.

[0112] Based on the response records of the over-limit test, specific evaluation metrics are constructed. For the active voltage over-limit test introduced above, specialized evaluation metrics are developed, including over-limit event response time, voltage recovery overshoot, success rate of regulating resource mobilization, and user-side device anomaly alarm rate. By comparing the dynamic response characteristics of the system before and after active over-limit testing, the rapid adjustment capability of the control strategy and the system resilience are quantified.

[0113] The second step involves calculating the corresponding revenue data based on grid operation data and user charging data before and after the power regulation command is executed, using a comprehensive cost-benefit analysis. The comprehensive cost-benefit analysis is shown in Formula 19. Formula 19; in, For benefit-cost ratio; Benefits from reduced network loss; This is due to changes in electricity purchase costs (usually a decrease). For the benefit of distributed power generation; Compensate electric vehicle users for costs; This refers to the system investment cost.

[0114] The user's subjective feelings are quantified into a satisfaction index through questionnaires. User satisfaction surveys and quantification involve designing questionnaires to quantify subjective feelings into a satisfaction index.

[0115] Calculate the peak-to-valley difference rate and load factor to assess the optimization of the load curve. The load curve optimization assessment calculates changes in indicators such as the peak-to-valley difference rate and load factor.

[0116] The third step involves using Monte Carlo simulation to randomly generate various test scenarios based on historical operating data and a library of typical scenarios to test the success rate of power regulation commands. Monte Carlo simulation tests randomly generate thousands of distributed power source and load scenarios to test the success rate of the control strategy.

[0117] Simulate extreme scenarios to verify system reliability. Extreme scenario stress testing simulates extreme weather, equipment failures, and other scenarios to verify system reliability. Parameter sensitivity analysis analyzes the impact of changes in key parameters on control effectiveness.

[0118] A social disturbance scenario was introduced to test the adaptability of power adjustment commands. In the social disturbance scenario test, a "social disturbance" scenario was introduced into the Monte Carlo simulation to simulate a sudden surge in charging demand caused by large-scale social events (such as music festivals or sporting events), or concentrated departure behavior caused by "charging anxiety" on social media platforms. The test aimed to assess the adaptability of the control strategy under atypical, high-uncertainty scenarios and evaluate the response effect of elastic resource scheduling under extreme social behavior.

[0119] The fourth step involves using historical operational data, association rule mining, cluster analysis, causal inference, and temporal clustering Hidden Markov Models to extract typical scenario libraries, control patterns, and user behavior evolution paths. The control strategy is then iteratively updated using reinforcement learning. Furthermore, online learning algorithms are used to update model parameters in real time and dynamically adjust the elastic contract incentive standards.

[0120] Among these, through knowledge accumulation and adaptive optimization, the control strategy is continuously improved based on operational data, thereby achieving the system's self-improvement.

[0121] First, we perform data mining and knowledge extraction to discover patterns from historical data and form reusable knowledge.

[0122] Association rule mining can employ Apriori or FP-Growth algorithms to discover association rules between distributed power output, voltage fluctuations, and optimal control strategies. Cluster analysis identifies typical scenarios by clustering historical operating scenarios and extracting a typical scenario library. Causal inference analysis uses a Pearl causal graph model to analyze the causal relationship between different control actions and voltage improvement effects. User behavior pattern evolution analysis uses temporal clustering and Hidden Markov Models to identify the evolutionary path of user charging behavior under long-term incentives, such as the shift from "rigid demand" to "flexible participation." The efficiency of different incentive strategies (price, social, points) on user behavior migration is analyzed to provide data support for dynamic adjustments to flexible contracts.

[0123] Next, the control strategy is updated online, automatically updating the control strategy and parameters based on new knowledge.

[0124] The policy iteration based on reinforcement learning is shown in Equation 20: Formula 20; in, Let be the state-action value function, representing the expected cumulative reward of taking action a in state s; The learning rate controls the degree to which new information affects the Q-value; Provides immediate rewards that reflect the effects of the current action; The discount factor balances the importance of immediate rewards versus future rewards; This is the optimal Q value under the next state s'.

[0125] Formula 20 learns better control strategies through continuous interaction with the environment.

[0126] Simultaneously, a digital twin simulation was set up to test the new strategy on the digital twin platform. After verifying its effectiveness, it was deployed to the physical system. An A / B testing mechanism was also implemented, with some nodes serving as the experimental group using the new strategy and the control group using the original strategy, to compare the effects.

[0127] Finally, the elastic resource potential is dynamically updated, and the elastic resource model is dynamically updated based on changes in user behavior.

[0128] Adaptive identification of user behavior patterns is implemented using online learning algorithms to update user behavior model parameters in real time. A dynamic adjustment mechanism for elastic contracts is established to dynamically adjust incentive compensation standards based on user participation and effectiveness. Adjustable potential prediction models are updated periodically with new data to improve prediction accuracy. Social influence-driven potential prediction updates are implemented, using user influence metrics on social networks (such as community activity, interaction frequency, and evaluation credibility) as new feature dimensions for elastic resource potential prediction, constructing a joint "behavior-social-potential" prediction model. Through an online learning mechanism, changes in users' charging willingness under social influence are captured in real time, dynamically adjusting their adjustable capacity and response probability to improve the accuracy of elastic resource scheduling.

[0129] 1. Improve the voltage safety and power quality of the distribution network: Through real-time sensing and evaluation and multi-scale optimization decision-making, the system can accurately predict the risk of voltage exceeding the limit and make advance or real-time adjustments using modeled elastic charging load resources.

[0130] Proactive prevention of voltage exceedance: Unlike traditional passive management, this application intervenes before or in the early stages of voltage deviation by adjusting the charging power (such as reducing the charging load of overvoltage nodes and increasing the charging load of low-voltage nodes) to stabilize the voltage within the qualified range, thereby significantly reducing the voltage unqualified factor (VUF) and the duration of exceedance.

[0131] Smoothing rapid voltage fluctuations: For voltage fluctuations on the order of seconds or minutes caused by photovoltaic cloud shadow effects, wind turbine turbulence, etc., a real-time feedback correction model and local fast response are used to offset the fluctuations through the millisecond-second adjustment capability of the charging load, effectively suppressing the amplitude of voltage fluctuations and improving power supply quality.

[0132] 2. Enhance the system's ability to absorb high proportions of renewable energy: Traditional distribution networks often have limited regulation capacity and need to adopt measures such as curtailing solar and wind power to ensure safety, resulting in the waste of clean energy. This application transforms the massive electric vehicle charging load into "virtual energy storage," providing bidirectional regulation space.

[0133] It provides flexibility in spatiotemporal shift: Through spatiotemporal elastic domain analysis, charging loads can be shifted from the off-peak hours of distributed power generation (such as at night) to the peak hours (such as at noon), directly absorbing excess photovoltaic power generation and reducing the impact of reverse power transmission on the upstream power grid.

[0134] Lowering the entry barriers and retrofitting costs for renewable energy access: Flexible load regulation to adapt to DG fluctuations can, to some extent, replace or delay traditional hardware upgrades (such as increasing transformer capacity, adding SVG, etc.) to enhance grid acceptance capacity, enabling more distributed photovoltaic and wind power to be economically and safely connected to the existing distribution network.

[0135] 3. Achieving economic operation of the power grid and win-win situation for all parties: By optimizing the model, pursuing economic optimization under the premise of ensuring safety, a new value stream is created.

[0136] Reducing overall system operating costs: Optimizing charging load scheduling can smooth out peak and valley loads, thereby reducing the procurement of high-priced electricity (lowering costs). At the same time, optimize power flow distribution and reduce network loss. Benefit-cost ratio (B / C) analysis can quantitatively verify its economic viability.

[0137] Creating benefits for users and improving satisfaction: Electric vehicle users participating in regulation can obtain economic returns through electricity price incentives (such as peak-valley price differences) or direct compensation. At the same time, the fairness-based allocation algorithm and user behavior modeling ensure flexible adjustments while meeting users' basic charging needs, protecting user experience and increasing user participation.

[0138] 4. Build a smart distribution network ecosystem with self-learning and self-adaptive capabilities: knowledge accumulation and adaptive optimization enable the system to go beyond the scope of static optimization and have the ability to continuously evolve.

[0139] Achieving self-iteration and optimization of control strategies: Through data mining and reinforcement learning, the system can automatically discover better control patterns from historical operating data, continuously update the strategy library and model parameters, and continuously improve the control effect over time.

[0140] Enhance the system's robustness to complex and ever-changing environments: Through robustness testing and dynamic model updates, the system can adapt to future uncertainties such as the growth of electric vehicle penetration, changes in user behavior patterns, and the access of new DG technologies, ensuring the long-term effectiveness of the method.

[0141] The solution incorporates unconventional or complementary / conflicting scenarios with conventional solutions, such as reverse scheduling, proactive limit testing, social behavior modeling, privacy protection collaboration, island autonomy, and user preference intervention, thereby enhancing its foresight, practicality, and system resilience.

[0142] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for autonomous grid operation based on flexible charging loads actively adapting to active distribution, characterized in that, include: Collect multi-source heterogeneous data, perform data fusion and power distribution network operation status estimation to obtain corresponding status estimates, and perform real-time voltage stability assessment and dynamic voltage stability index assessment based on the status estimates. Based on the user-side load data and external environment information contained in the multi-source heterogeneous data, the charging load regulation capability is quantified through spatiotemporal prediction, virtual energy storage aggregation and elastic domain analysis. Based on the state estimate, the voltage sensitivity matrix in the dynamic voltage stability index, the aggregated adjustable capacity of the virtual energy storage, and the migration range in time and space corresponding to the adjustment capability of the charging load, the power adjustment command of each cluster corresponding to the virtual energy storage aggregation is generated through day-day-intraday-real-time multi-timescale collaborative optimization decision-making. The power adjustment commands for each cluster are decomposed and sent to each charging pile so that the local controller can adaptively execute the power adjustment commands and monitor and correct the execution deviation in real time. Based on the measured voltage data of each charging pile during the actual execution of the power adjustment command, a multi-dimensional quantitative evaluation and verification are conducted, and the control strategy is continuously improved based on the operational data.

2. The method according to claim 1, characterized in that, Collect multi-source heterogeneous data, perform data fusion, and estimate the operating status of the distribution network to obtain the corresponding state estimates, specifically including: The system collects grid operation parameters based on synchronous phasor measurement units, user-side load data based on smart meters, and distributed power source data based on distributed power source monitoring systems to obtain multi-source heterogeneous data. For the aforementioned multi-source heterogeneous data, a credibility-weighted fusion is performed to obtain fused data; Based on the weight diagonal matrix corresponding to the credibility weight, the measurement equation is established by weighted least squares method to obtain the corresponding optimal state variables; Based on the measurement function corresponding to the optimal state variable, the corresponding state estimate is obtained by recursive estimation through Kalman filtering dynamic estimation.

3. The method according to claim 1, characterized in that, Based on the state estimate, real-time voltage stability assessment and dynamic voltage stability index assessment are performed, specifically including: Real-time voltage stability assessment is performed. Based on the values ​​in the state estimate, the corresponding Jacobian matrix is ​​determined. Based on the Jacobian matrix, the active power-voltage sensitivity matrix and the reactive power-voltage sensitivity matrix are obtained. Based on the active power-voltage sensitivity matrix and the reactive power-voltage sensitivity matrix, the voltage sensitivity matrix is ​​obtained. The dynamic voltage stability index is evaluated based on the voltage value in the state estimate. It allows voltage to perform over-limit tests within a set range and monitors the rate of change of voltage at key nodes and load response characteristics in real time during the over-limit process.

4. The method according to claim 1, characterized in that, Based on the user-side load data and external environmental information contained in the multi-source heterogeneous data, the charging load regulation capability is quantified through spatiotemporal prediction, virtual energy storage aggregation, and elastic domain analysis, specifically including: Based on the user-side load data and external environment information contained in the multi-source heterogeneous data, user charging behavior is modeled through spatiotemporal prediction. The individual charging needs corresponding to the user charging behavior model are aggregated to obtain multiple clusters and corresponding virtual energy storage. Based on the user charging behavior modeling, elastic domain analysis is performed for each cluster to quantify the migration range of charging load in time and space. Based on the spatiotemporal distribution of the charging load corresponding to the user charging behavior model, the aggregate adjustable capacity of virtual energy storage, and the migration range of the charging load in time and space, the adjustment capability of the charging load is obtained.

5. The method according to claim 4, characterized in that, Based on the user-side load data and external environmental information contained in the multi-source heterogeneous data, user charging behavior is modeled through spatiotemporal prediction, specifically including: Based on the user-side load data and external environment information contained in the multi-source heterogeneous data, a spatiotemporal prediction model is established through deep learning to predict the spatiotemporal distribution of charging load. Based on user location, battery state of charge, dwell time, and electricity price information from the external environment, a Markov decision process model is used to obtain the corresponding user sequence decision; and the user sequence decision is then corrected based on social attributes. Based on the changes in user charging power and the changes in electricity prices in the external environment, the corresponding demand elasticity coefficient is obtained.

6. The method according to claim 5, characterized in that, The individual charging demands corresponding to the user charging behavior model are aggregated to obtain multiple clusters and corresponding virtual energy storage, specifically including: For each user's sequence decision-related individual charging demand, the charging load is divided into multiple clusters using a clustering algorithm; Based on the cluster partitioning results, the aggregate adjustable capacity is calculated; the aggregate adjustable capacity includes the upper and lower limits of aggregate power, the upper and lower limits of aggregate energy, and the energy time-series recursive relationship; Identify the virtual energy storage characteristic parameters of each cluster using historical data; For bidirectional charging and discharging scenarios, the discharge model is extended.

7. The method according to claim 6, characterized in that, Based on the user charging behavior modeling, elastic domain analysis is performed on each cluster to quantify the temporal and spatial migration range of the charging load, specifically including: Based on the individual charging needs, a feasible charging spatiotemporal domain is constructed; Based on the cluster partitioning results, a flexible resource map is drawn on the geographic information system platform to show the distribution of adjustable charging capacity in different regions and time periods. The potential for improvement of voltage distribution by spatiotemporal migration is quantitatively evaluated through heuristic algorithms. Construct a coupling matrix for inter-regional charging load migration to assess the feasibility and cost of transferring some charging demand to neighboring regions.

8. The method according to claim 1, characterized in that, Based on the state estimate, the voltage sensitivity matrix in the dynamic voltage stability index, the aggregated adjustable capacity of the virtual energy storage, and the temporal and spatial migration range corresponding to the charging load's adjustment capability, power adjustment commands for each cluster corresponding to the virtual energy storage aggregation are generated through day-day, intraday, and real-time multi-timescale collaborative optimization decision-making. Specifically, this includes: Based on the aggregate adjustable capacity of the virtual energy storage and the migration range in time and space corresponding to the adjustment capability of the charging load as constraint boundaries, a day-ahead stochastic optimization model is constructed, and a baseline scheduling plan is formulated; an intraday rolling optimization model is formulated through a model predictive control framework; a real-time feedback correction model is constructed to adjust power based on real-time voltage deviation; and a reverse scheduling scenario is extended to proactively increase the charging load to absorb renewable energy when the output of distributed power sources is excessive. Based on the voltage sensitivity matrix in the dynamic voltage stability index, the corresponding upper and lower voltage limits are transformed into feasible domain boundaries for charging power adjustment, and a time-varying voltage safety boundary is constructed; and a constraint relaxation and penalty mechanism is introduced. Based on the cluster partitioning results of the virtual energy storage, the global optimization problem is decomposed into subproblems of each cluster using the alternating direction multiplier method for parallel solution; and the consensus optimization algorithm is used to exchange boundary information among the cluster controllers to iteratively achieve the global optimum; and a federated learning mechanism is introduced to protect user privacy by ensuring that each cluster does not share original user data and only uploads model gradients or optimization boundary parameters.

9. The method according to claim 1, characterized in that, The power adjustment commands for each cluster are decomposed and sent to each charging pile. The local controller adaptively executes the power adjustment commands and monitors and corrects execution deviations in real time. Specifically, this includes: For the power adjustment commands of each cluster, the total adjustment amount of the aggregation layer is decomposed into the single power command of each charging pile through a power allocation algorithm based on fairness, and sent to each charging pile through an encryption algorithm; the elastic weight is dynamically adjusted according to the user's active settings; and a multi-level priority strategy is designed to handle the conflict between emergency control and regular scheduling. The local controller, based on the issued power command of the single pile and the locally collected user target state of charge, performs local optimization through model predictive control to obtain the corresponding power data; and automatically switches to islanded autonomous mode when communication is interrupted, and performs power self-balancing based on local voltage-frequency droop characteristics and energy storage status. The local controller executes locally optimized power data, detects the deviation between actual power and commanded power, and triggers adaptive correction when the deviation exceeds the allowable error threshold. It also handles communication interruptions and equipment failures according to the preset abnormal situation handling plan.

10. The method according to claim 3, characterized in that, Based on the measured voltage data of each charging pile during the actual execution of the power adjustment command, a multi-dimensional quantitative evaluation and verification are conducted, and the control strategy is continuously improved based on the operational data, specifically including: Based on the measured voltage data of each charging pile during the actual execution of the power adjustment command, the voltage deviation index, voltage qualification rate, and the reduction ratio of voltage fluctuation rate before and after control are statistically analyzed to identify and improve the corresponding nodes; and based on the response records of the over-limit test, a dedicated evaluation index is constructed. Based on the power grid operation data and user charging data before and after the execution of the power adjustment command, the corresponding revenue data is calculated through comprehensive cost-benefit analysis; user subjective feelings are quantified into a satisfaction index through a questionnaire; and the peak-valley difference rate and load factor are calculated to evaluate the optimization degree of the load curve. Based on historical operating data and a library of typical scenarios, multiple test scenarios are randomly generated through Monte Carlo simulation to test the success rate of the power regulation command; extreme scenarios are simulated to verify the system reliability; and a social disturbance scenario is introduced to test the adaptability of the power regulation command. Based on historical operational data, association rule mining, cluster analysis, causal inference, and temporal clustering Hidden Markov Model are used to extract typical scenario libraries, control laws, and user behavior evolution paths. The control strategy is iteratively updated through reinforcement learning. The model parameters are updated in real time and the elastic contract incentive standard is dynamically adjusted through online learning algorithms.

Images